nvidia notes
This commit is contained in:
parent
2653d33292
commit
1ae0e88627
19
docs/nvidia_notes
Normal file
19
docs/nvidia_notes
Normal file
|
@ -0,0 +1,19 @@
|
|||
A100 has 432 tensor cores
|
||||
|
||||
8x4x8 FP16 matmul = 256 FMAs
|
||||
8x4x4 TF32 matmul = 128 FMAs
|
||||
|
||||
432 * 256 * 1410 mhz * 2(m and a) = 312 TOPS
|
||||
|
||||
Why aren't the other accelerators 3D like this?
|
||||
|
||||
--
|
||||
|
||||
SNPE is using 4x4x4 -> 4x4 (64 FMAs) in the convs.
|
||||
Then it's accumulating in that matrix.
|
||||
|
||||
256 ALUs (1 FMA per cycle)
|
||||
652 mhz
|
||||
---
|
||||
319 GFLOPS
|
||||
|
Loading…
Reference in a new issue